Trish Whetzel

I built KidSketch for the Gemini Live Agent Challenge - an AI agent that turns children's drawings into illustrated story movies #GeminiLiveAgentChallenge

🚨 Anthropic just showed a 24-minute workshop on how to actually do prompts for Claude. Taught by the people who built it. Free. No registration. No paywall. I've seen $300 courses that don't cover what they teach in the first 8 minutes. Watch it and bookmark it now.

@nlycskn @marktechson I'm excited and grateful to be there! Definitely want to catch up with you 😊

#GoogleIO is in less than a week! Who will be there? 🙌

Meet the 100+ pioneering startups joining the second Gemini Startup Forum. 🌎🛠 On June 16–17 at Google’s Sunnyvale HQ, these global founders will take part in an intensive, two-day summit designed to accelerate the global scale and impact of their AI solutions. Participants will collaborate directly with experts from across @Google to overcome technical hurdles, refine growth strategies, and scale their solutions globally using Google’s premier AI infrastructure. 🚀 Learn more about the cohort: goo.gle/48US2BM

@rseroter Great, I see the Plan mode and am testing it out now for a simple web site. For "ask", yes to ask a specific question about line(s) of code or a general coding question. I use these modes often in Cursor and miss these in Antigravity.

@TrishWhetzel What would you like to see here? We've got a basic plan mode (antigravity.google/docs/agent-mod…), but are you also looking for a one-off "ask" interface for single-turn conversations? Tell me more!

Google Antigravity beats Claude at coding—but only if you stop acting like a programmer howtogeek.com/google-antigra… < all these tools are great; just pick one. But regardless, don't bring your old approach to new tools.

Calling all developers 📣 Sign up to participate in user research studies and influence how Google evolves the tools, APIs, and guidance for millions of people around the world. 🌎 Get started by telling us more about your development experience: goo.gle/4uPE4cU

Go beyond simple chatbots and build a distributed multi-agent system—this codelab will show you how → goo.gle/4udm8Jc Build the architecture needed to enable multiple agents to work together to achieve a shared goal.

Forward Deployed Engineers shouldn't *be* your AI implementation strategy. But they can be an important piece. At @googlecloud, it's game on. The best in the biz is building out our team so we can help enterprises do AI well at scale. Apply! linkedin.com/posts/rarni_go…

Cool Ambassador programs open globally: - OpenAI Codex Ambassadors Program: developers.openai.com/community/code… - Microsoft Student Ambassadors: learn.microsoft.com/en-us/training… developers.openai.com/community/code…

@googledevs Are these sessions presented by Google Dev Rel or community presentations by GDEs?

Master the tools to build your own grounded AI companion with the experts who use them every day. Register 💻 → goo.gle/4npl2Hu Join our live AI Learning Labs for a hands-on deep dive into: 🔹 Gemini CLI & Python 🔹 Agent Grounding via Google Search 🔹 MCP Server & Agent ADK 🔹 Custom Avatar generation with Imagen Move past the basics and build a one-of-a-kind interactive agent with real-time support from Google Developer Experts. 📅 May 27-29

Get @GoogleCloud certified. ☁️ Google Cloud customers can join GEAR to apply and access an up to 9-week cohort featuring technical mentorship and instructor-led training into various topics → goo.gle/4wqeCfF

If I had 6 months to become an Applied AI Engineer. I’d do this. Stage 1: Python + Production APIs FastAPI, async, error handling, webhooks, REST/GraphQL, third-party SDKs. Stage 2: LLM Fundamentals for Production Tokens, context windows, model routing, embeddings, cost/latency tradeoffs. Stage 3: Prompt Engineering + Structured Outputs System prompts, few-shot chains, Pydantic/JSON validation, prompt versioning, unit evals. Stage 4: RAG + Knowledge Grounding Chunking strategies, hybrid search, rerankers, vector DBs, metadata filtering, citation tracking. Stage 5: AI Workflows + Orchestration Tool calling, state machines, human-in-the-loop, retry logic, fallback chains, session memory. Stage 6: Build Production-Ready Apps Domain-specific copilots, automation pipelines, streaming UIs, graceful degradation, rate limiting. Stage 7: Evaluation + Reliability Accuracy scoring, hallucination detection, RAGAS/DeepEval, regression testing, A/B output validation. Stage 8: AI Infrastructure + Optimization vLLM, Ollama, quantization, KV caching, response streaming, token cost tracking, edge deployment. Stage 9: Deployment + Observability Docker, CI/CD, cloud hosting, distributed tracing, structured logging, alerting, canary releases. Stage 10: AI Security + Guardrails Input/output filtering, prompt injection defense, PII redaction, compliance checks, sandboxing. Stage 11: Open Source + Portfolio Ship end-to-end apps publicly, write architecture docs, record demo walkthroughs, publish eval reports. Stage 12: Apply Applied AI Engineer, GenAI Developer, AI Integration Engineer, LLM Solutions roles. Most people stay stuck watching tutorials. Builders get hired.

Come celebrate after Google IO with the DeepMind team in SF on May 21st :) Lots to look forward to!! Link in 🧵

Choosing between APIs, MCP, and Agent-to-Agent architectures lakshmanok.medium.com/choosing-betwe… < @lak_luster says to default to function tools that use REST APIs, and only bring in MCP or A2A when there's more reasoning involved, or dynamic use cases.

The Google for Startups AI Agents Challenge is now LIVE. 🦾⚡️ Build autonomous systems with @GeminiApp and the recently upgraded Agent Development Kit for a chance to win a share of $90,000 and $500 in @GoogleCloud credits! Register now to start building: goo.gle/4cpi2pB

Looking forward to this hackathon!

Claude Code ships with 5 architectural layers most engineers never open. Not features. Not settings. Layers — each solving a distinct problem that LLMs alone can't solve. And four of them have nothing to do with prompting. Here's the full Agent Development Kit: Layer 1 — CLAUDE.md → The Memory Layer Architecture rules, naming conventions, test expectations, repo map. Always loaded. Always active. Two scopes: • ~/.claude/CLAUDE.md → global • .claude/CLAUDE.md → project This isn't context you paste in before every session. It's context that never needs repeating. The agent's constitution. Layer 2 — Skills → The Knowledge Layer Each SKILL.md carries a description. Claude matches it at runtime and forks the skill into an isolated subagent. On-demand, never always-on. Task-specific knowledge without inflating your main context window. Modular by design. Layer 3 — Hooks → The Guardrail Layer PreToolUse → PostToolUse → SessionStart → Stop → SubagentStop This is the layer most teams skip. And the one they regret skipping first. Hooks are NOT AI. They're deterministic event-driven shell commands. • Auto-lint on every Write • Hard-block on rm -rf • Slack notification on Stop Event fires → Matcher checks → Command runs Quality enforced at the infrastructure level. Not the prompt level. Layer 4 — Subagents → The Delegation Layer Each subagent gets its own context window, model, tools, and permissions. Main agent delegates down. Receives results up. That's it. No infinite recursion — subagents can't spawn subagents. Main context stays clean. Hard boundaries by design. Layer 5 — Plugins → The Distribution Layer Bundle your skills + agents + hooks + commands into a plugin. One install. Whole team inherits the behavior. Think npm packages — but for what your agent knows how to do. Wrapping everything: → MCP Servers on the left (GitHub, databases, APIs, custom integrations) → Agent Teams on the right (parallel execution, message passing, shared permissions) The 5-layer stack in one line: CLAUDE.md sets rules → Skills provide expertise → Hooks enforce quality → Subagents delegate work → Plugins distribute to the team Most production failures in agentic systems trace back to one missing layer. Which one is the gap in your current setup?

Scion is a new multi-agent orchestration tool that orchestrates agents (Claude Code, Gemini CLI, Codex, and others) as isolated, concurrent processes. Each agent gets its own container, git worktree, and credentials — so they can work on different parts of your project without stepping on each other. github.com/GoogleCloudPla…

🪨 New @Kaggle competition (with a $50,000 prize!) just dropped, and it's near to my geophysicist heart: Drilling a horizontal well is like navigating underground without a map. The path forward runs through layers of rock you can’t see. Build models to predict the geology along a horizontal wellbore. Your work will help automate and improve drilling operations in the oil and gas industry.

May the Code be with you! 🛠️✨ Build the future of the galaxy at #GoogleIO on May 19–20. Set your coordinates ➡️ goo.gle/io2026-1